Anthropic launches its Cyber Mission
On 8 October 2026 Anthropic announced the Anthropic Cyber Mission, a long-term effort to give defenders tools, research and resources, and began it with two programmes. The Critical Infrastructure Defense Program brings Claude models, on-site engineers and threat research to eleven founding partners that secure operational technology. OSS Scanner gives open-source projects free, periodic scans by Anthropic's strongest models. The scanner's reports are model-generated and sent without human review; Anthropic expects a true-positive rate above 90%.
Why it matters
Anthropic had been giving selected defenders access to its models (Project Glasswing, the Cyber Verification Program). It now runs programmes of its own: engineers placed with the firms that patch power, water and transport systems, and a scanner that sends maintainers unreviewed findings because, it says, human triage cannot keep up. The company forecasts that in two years AI will favour defence, but warns that in the near term it may not: exploiting flaws has become cheaper, while verifying and fixing them is still slow.
What the pages say. The announcement of 8 October names two starting areas. The Critical Infrastructure Defense Program (CIDP) targets the operational technology behind power grids, water systems and transport networks, and government systems. Its founding partners are Accenture, Booz Allen, CrowdStrike, Deloitte, Dragos, Hitachi, Insane Cyber, Nozomi Networks, Palo Alto Networks, PwC and Rockwell Automation. The page says the work is under way, that several partners are already using Claude to fix vulnerabilities, that the first step is a small cohort of providers, and that critical infrastructure is hard to defend "in many ways that AI cannot fix". It adds that since June, when it launched a defence programme for state, local, tribal and territorial governments, Anthropic has offered models and technical support to "more than half of all US states". OSS Scanner is modelled on Google's OSS-Fuzz. Enrolled projects receive periodic scans free of charge; each report carries a proof of concept, an explanation and, where one exists, a suggested fix. The page says reports are model-generated and sent without human review, so some will contain errors such as a wrong severity rating, and that Anthropic "expect[s] a true-positive rate above 90%". The Frontier Red Team's post of the same day adds: over 29,000 candidate vulnerabilities found in six months, of which about 6,000 had been reviewed by people; nearly 5,000 reports already sent to maintainers on request without validation; and a check in which Anthropic's penetration testers examined 97 critical and high-severity findings across 48 projects, of which 85 (88%) met the bar of its disclosure process, 11 were real but duplicated known issues and one was a false positive. The scanner runs on the company's strongest models (the post names Claude Mythos); projects enrol by a pull request to a GitHub repository and are admitted case by case on criteria like OSS-Fuzz's. Money and context. The page says Anthropic has funded the organisations behind widely used open-source code (the Python Software Foundation, Alpha-Omega and OpenSSF through the Linux Foundation, the Apache Software Foundation) and the Defender Advantage Fund (0xDAF), launched in August, which supports pilots and keeps OSS Scanner free; no amounts are given. In Glasswing, months often passed between a vulnerability being found and fixed, and for operational technology a fix may in rare cases wait for decades until it can be applied safely to running machinery. Earlier that week Anthropic had merged Project Glasswing into an expanded Cyber Verification Program. What the record does not claim. The 90% rate is an expectation, and the 88% comes from Anthropic's own check of a sample of its own findings; no assessment of the service by maintainers or outside evaluators was read. The partners' quotations on the page are endorsements and were not used. The page does not say how many partners will join or when, and gives no cost for the programme. The forecast that AI will favour defence in two years is Anthropic's own.